Today, vast amounts of protein-small molecule binding sites are available in the Protein Data Bank (PDB). Comprehensive comparisons are computationally demanding, but will be useful for prediction of protein functions and drug discovery. We proposed a tremendously fast algorithm called 'SketchSort' that makes it possible to enumerate similar pairs in a huge number of binding sites. We applied the proposed method to all-pair similarity searches for 5.5 million known and potential binding sites , and discovered over 49 million similar pairs of binding sites.